Pedestrian safety is a topic that concerns everyone. Not only does it concern researchers and practitioners who dedicate themselves to improve road safety, it also concerns all road users, as everyone is a pedestrian at some point in the transportation system. The high frequency of pedestrian crash injuries and the great possibility of fatal consequences have made pedestrian safety a great focus in road safety research. Pedestrian safety becomes even more of a problem at non-signalized locations when compared to signalized crosswalk locations, due to the absence of traffic lights controlling the traffic. Different methods have been proposed and empirical studies have been conducted to investigate pedestrian safety. Despite the extensive literature on investigating pedestrian safety at non-signalized crosswalk locations, much remains to be done, especially in studying interactions between the pedestrian and vehicles and their behavior during those interactions. Therefore, this dissertation aims to improve data collection methods for pedestrian safety analysis and to develop a methodological framework to investigate pedestrian safety at non-signalized crosswalk locations and implement such a framework using video data collected from different crosswalk locations with the help from vision-based tracking technology.
The work of the dissertation started with reviewing methodologies and data collection methods in previous studies. Methods used in past studies were classified into five different approaches. These are the crash data approach and four surrogate safety approaches, namely, the traffic data approach, the conflict event approach, the behavioral analysis approach, and the perception analysis approach. Issues in the use of terms and definitions, methodologies applied, and data used in previous studies were summarized. Some preliminary data collection work had indicated the limitations of using regular visible spectrum cameras in low visibility conditions. To overcome the limitations that regular visible spectrum cameras have encountered during the data collection process, the thermal camera was introduced and its performance in road user detection, classification, and speed measurement was validated through its comparison to the use of the regular camera. Validation results showed an evidently better performance from thermal camera for low visibility and shadow conditions, particularly when tracking pedestrians and cyclists. However, the regular camera narrowly outperformed the thermal camera during daytime. For speed measurements, the thermal camera was consistently more accurate than the regular camera at daytime and nighttime. To evaluate existing measures in investigating pedestrian-vehicle interactions at non-signalized crosswalk locations, a study was conducted to investigate pedestrian safety at nighttime. Although, the methodology applied in the study performed well in looking at pedestrian-vehicle interactions, further limitations of using safety measure methods were discovered upon the completion of the study. A novel framework, which evaluates pedestrian safety by looking at the interaction between the pedestrian and the vehicle, and their behavior during the interactions, was proposed and illustrated through a case study. The framework was further tested through a study to compare the performance of three main non-signalized crosswalk types, including uncontrolled, marked, and stop sign controlled crosswalks, on pedestrian safety using data collected from different sites in Montreal. Among the three types of non-signalized crosswalks, stop sign controlled crosswalks had the best performance in protecting pedestrians while uncontrolled crosswalks performed the worst. To explore the extensive applications of the framework, the investigation of cyclist-pedestrian interactions was introduced as it has been a major road safety problem but underestimated in previous research. Marked crosswalks alone fail to protect pedestrians from passing cyclists. Besides, pedestrian safety at crossings on cycling facilities with downhill grades was found to be a great issue.
In brief, the dissertation will: 1) provide a comprehensive literature review that acts as a practical reference to investigating pedestrian safety at non-signalized crosswalk locations, 2) introduce a promising alternative, the use of the thermal camera, to overcome the limitations of using the visible spectrum camera for automated traffic data collection, 3) propose a new framework that describes pedestrian-vehicle interactions more precisely, compared to previous studies. This framework is promising for different purposes in road safety on various topics, such as the analysis of interactions between different types of road users, the simulation of road user interactions, validations of safety treatments, and the performance evaluations of autonomous vehicles.